The reliability of a 13-item questionnaire designed to assess the humanistic behaviors of internal medicine residents and the reliability of nurses as raters of those behaviors were examined. Twenty-five residents were evaluated by 10 or 11 nurses on two general medicine services and on cardiology and hematology-oncology services in a large, highly specialized department of internal medicine. Using an application of generalizability theory, which extends beyond classical test theory to establish estimates of multiple-error sources, the investigators calculated reliability-like coefficients for each of the services. The coefficients were .95 and .85 for the two general medicine services, .67 for cardiology, and .88 for hematology-oncology. These findings indicate that the questionnaire is a reliable instrument for assessing humanistic behavior and identifying reliable raters in groups of nurses.